Why Professionals Struggle to Trust AI
Hello, I am
Yuki Fukiage, CEO of Lean Stack Inc..
Introduction
In this article, as someone who promotes AI, I will dare to make a bold statement.
It is correct for professionals to be unable to trust AI.
This is not just me talking my own book; I genuinely believe this.
They deal with the law.
They create documents that influence people's lives.
They work in a world where not a single character can be wrong.
There is no way such people would simply say "Yes, let's leave it to AI!" when told to do so.
In fact, it would be scarier if they trusted it easily.
However, "being unable to trust" and "not using" are completely different things.
In this article, I will fully validate the reasons why professionals cannot trust AI, and then talk about how we are still overcoming that barrier.
The reasons why professionals cannot trust AI are all valid
I am sure there are some professionals who have tried throwing a draft of work rules into ChatGPT.
What comes back is a 60-70 point answer that is "well, it's not wrong, but..."
As a general theory, it sounds plausible.
But when it comes to whether it reflects the industry, scale, and regional customs of their clients, it does not at all.
For professionals, 70 points can be the same as 0 points.
This is not an exaggeration at all.
There is no such thing as "mostly correct" in law.
If you get one article of the Labor Standards Act wrong, or one requirement for social insurance application wrong, it will cause actual harm to your client.
Ask GPT, confirm with Claude, cross-check with Gemini—
—in the end, they are all just "averagely smart" and do not take care of the detailed conditional branching of the law.
What is the maximum number of hours for the special provisions of the Article 36 agreement in this industry?
Does this employee qualify for the expansion of social insurance coverage? Their annual income exceeds 1.06 million yen, but their weekly scheduled working hours are...
General-purpose AI cannot properly answer these questions that fly around in the field of practical work every day.
Therefore, feeling that AI is useless is completely correct.
In fact, when I talk to professionals about AI, about 80% of them say, 'AI is stupid, isn't it?'
As you say, it is undeniable that this happens in highly specialized fields.
'If that's the case, I'll do it myself' is the most dangerous approach
I can't trust AI.
General-purpose AI is useless.
If so, I should just do it myself.
It's more reliable if a person does it.
——This judgment sounds logical.
But this is the most dangerous judgment of all.
What awaits you if you keep choosing to 'do it yourself'?
Personal dependency. The head of the office ends up carrying everything and nearly collapses during the busy season every month.
Labor shortages. Even if you try to hire, young people don't come to professional firm job postings. Even if they do, they don't stay.
Depletion of time. Chasing after client support, you can't do new sales, service improvements, or anything else.
Everything is an extension of the judgment that 'it's more reliable if a person does it,' isn't it?
Not trusting AI and ignoring the existence of AI are completely different things.
You don't have to trust it.
But I want you to realize that the choice to 'keep doing it all by yourselves' is quietly strangling your firm.
'It's working for now'—have you ever calmly considered how long that 'now' will last?
How we are exceeding the limits of general-purpose AI
Having read this far, some of you must be thinking, "So, what should I do?"
I will give you the answer.
Stop trying to force general-purpose AI to do the impossible.
GPT, Claude, and Gemini are certainly impressive.
However, they are ultimately just "averagely smart" AIs.
In fields where "accuracy in the details" is life—such as legal statutes, notifications, judicial precedents, and actual administrative practices—general-purpose AI has its limits.
Refining prompts makes it a bit better. Incorporating knowledge makes it better. But "a bit better" isn't enough for professional work. There are many professionals out there holding their heads in their hands like this...
So, what do we do?
What we are doing is using AI that allows for detailed rule configuration.
It's not that GPT or Gemini are bad.
In fact, ChatGPT has GPTs and Gemini has Gems, which are features that allow you to accumulate knowledge and customize them.
You can certainly load "a certain level of expertise" into them.
But there is a massive gap between "a certain level" and "a level where you can entrust legal work."
Professional work is, after all, a collection of rules.
Detailed conditional branching in laws. Exception clauses for each industry. A series of judgments like "in this case, do this, but if these conditions overlap, do that."
How finely and accurately you can set these rules for the AIis exactly where the battle is won.
The reason general-purpose AI stops at 70 points isn't because the AI isn't smart. It's because it isn't designed to master rules.
Conversely, if you choose an AI that allows for detailed rule configuration, you can load your firm's expertise directly into the AI.
Many people stop at "I asked ChatGPT and it didn't work," but that is just a problem of how you choose your tools.
Choose an AI that can enhance your expertise.
Thoroughly instill your own rules into the AI.
That alone will overturn the premise that 'AI is useless'.
There is no need to "trust" AI
Even after reading this far, some people might still be thinking, 'But is it really okay?'
That is perfectly fine.
In fact, you don't have to trust it.
What is the greatest strength of a professional?
It is the ability to 'doubt, confirm, and verify'.
Professionals who deal with the law always check the grounds.
Consult the articles of the law.
Read the notifications.
Research legal precedents.
You just need to apply that same power to AI output.
There is no need to 'rely' on AI.
You just need to 'verify' the AI's output.
Not reliance, but verification.
This is the correct way for professionals to engage with AI.
And you know, once you verify it, you will realize something.
'Wait, this is overwhelmingly faster than doing it from scratch myself.'
Since you don't have to trust it,
I want you to try it once and verify it once.
That is all.
Summary
It is correct for professionals to be skeptical of AI.
It is only natural that those who perform highly specialized work feel a sense of discomfort with 'average-scoring AI'.
However, 'not using it because you don't trust it' is a mistake.
The limitations of general-purpose AI are real.
But AI that transcends those limitations already exists.
You don't have to trust it.
But it is a waste not to try it.
Will you stand still while remaining skeptical, or will you move forward without needing to trust it?
The choice is yours.
Thank you for reading until the end.
'I want to try using AI in my firm,'
'I want to consult on which AI is suitable for professionals,'
'I can't trust AI, so I want to check the accuracy of its output! (We show this for free)'
If you are interested, please feel free to consult with us via the official LINE below.
Click here for LINE consultation!
↓
https://lin.ee/I73KUiN
We also share information about AI and DX on YouTube.
↓
https://youtube.com/channel/UC-f1ZI9MiME-GC0l1gwNf4w?si=WeyqGjBSvVXwSaDB